How Many AI Agents Do You Have?

There's a question showing up in pitch rooms right now that's making agency people twitch: "How many AI agents does your company have?"

Digiday reported it this month -- clients opening AI conversations with a headcount question for software. Agency executives hate it. One called automation-at-face-value a race to the bottom. The consensus is that clients are asking the wrong question and someone should teach them a better one.

The question isn't naive. It's diagnostic. And what it's diagnosing is us.

Clients learned to ask it because we taught them to. For two years the industry led every pitch with inventory -- the stack slide, the agent engine, the platform partnerships, the proprietary tool with the not-so-clever name. They marketed what was countable, so clients are auditing the count. You can't spend two years telling the market your differentiator is a number and then act insulted when the market asks for the number.

Here's the problem: anything countable is in-houseable. A seat has a price. A tool has a license. An agent has a monthly cost that procurement can line up against your retainer in one spreadsheet. The moment your differentiator has a number next to it, you've handed the client two documents at once -- a shopping list and a budget line.

I've watched it happen. The capabilities slide goes up, and somewhere in the room a pen comes out. Not to take notes on your thinking. To mark the rows worth pricing without you. That's not bad faith. Procurement's entire job is converting things into units -- it's the job description, not a betrayal. Every unit you show them, they will eventually price on their own.

Which means the agencies losing this pitch are the ones answering that question.

The better move is the follow-up almost nobody makes: what do we sell that has no unit? Judgment doesn't have a count. Taste doesn't have a count. Knowing which problem deserves the tool -- and when to leave the tool in the drawer -- has never appeared on a procurement sheet, because there's no cell for it. There's no per-seat license for the person who reads the brief and says "this is the wrong brief." No monthly cost for the call that kills a bad idea before it burns six weeks of production. That's the thing worth defending, and it's exactly the thing the stack slide buries.

It's why Sageworx doesn't lead with tool inventory. The inventory isn't the product. Curation is -- which member, on which problem, making which call, and knowing when the right answer is no tool at all. A client can replicate a stack from a demo. They can't replicate the judgment about what the stack is for.

So the next time the agent-count question lands, understand what's actually happening in the room. When a client asks how many AI agents you have, they're not evaluating you. They're pricing your replacement.

Answer with a number and you've confirmed the estimate. Sell the thing that has no price instead.